ABSTRACTThere has recently been an upsurge in interest in variation in both synchronic and diachronic linguistics. This paper begins with two hypotheses: first, that phonologists ought these days to be interested in how variation is analysed, and secondly, that the Standard Generative approach involving underlying identity is not the way to do it. It is shown that a particular difference between RP and Standard Scottish English can be appropriately analysed within a Lexical Phonology allowing underlying discrepancies between dialects. However, the introduction of underspecification into the model allows the (badly motivated) reintroduction of underlying identity. The paper concludes with a survey of further problems with underspecification theory.